Model for exploiting associative matching in AI production systems

摘要

A content-addressable model of production systems (CAMPUS) has been developed. The main idea is to achieve high execution performance in production systems by exploiting the potential fine-grain data parallelism. The facts and the rules of a production system are uniformly represented as content-addressable memory (CAM) tables. CAMPUS differs from other CAM-inspired models in that it is based on a non-state-saving and “lazy” matching algorithm. The production system execution cycle is represented by a small number of associative search operations over the CAM tables. The number does not depend, or depends slightly, on the number of the rules and the number of the facts in the production system. The model makes possible efficient implementation of large production systems in fast CAM. An experimental CAMPUS realisation of the production language CLIPS is also reported. The production systems execution time for a large number of processed facts is about 1000 times lower than the corresponding CLIPS execution time on a standard computer architecture.